Advanced Moment Estimation & Optimization with …past.rinfinance.com/agenda/2015/workshop/RossBennett.pdf · 2015-06-08 · PortfolioAnalytics is an R package designed to provide

Advanced Moment Estimation &Optimization with PortfolioAnalyticsR/Finance 2015

Ross Bennett

Advanced Moment Estimation & Optimization with PortfolioAnalytics

1 of 38

OverviewDiscuss Portfolio Optimization

Introduce PortfolioAnalytics

Demonstrate PortfolioAnalytics with Examples

·

·

·

2/38


2 of 38

Modern Portfolio Theory"Modern" Portfolio Theory (MPT) was introduced by Harry Markowitz in 1952.

In general, MPT states that an investor's objective is to maximize portfolio expected return for agiven amount of risk.

General Objectives

How do we de�ne risk? What about more complex objectives and constraints?

Maximize a measure of gain per unit measure of risk

Minimize a measure of risk

·

·

3/38


3 of 38

Portfolio Optimization ObjectivesMinimize Risk

Maximize Risk Adjusted Return

Risk Budgets

Maximize a Utility Function

·

Volatility

Tail Loss (VaR, ES)

Other Downside Risk Measure

-

-

-

·

Sharpe Ratio, Modi�ed Sharpe Ratio

Several Others

-

-

·

Equal Component Contribution to Risk (i.e. Risk Parity)

Limits on Component Contribution

-

-

·

Quadratic, CRRA, etc.-

4/38


4 of 38

PortfolioAnalytics OverviewPortfolioAnalytics is an R package designed to provide numerical solutions and visualizationsfor portfolio optimization problems with complex constraints and objectives.

Support for multiple constraint and objective types

Modular constraints and objectives

An objective function can be any valid R function

Support for user de�ned moment functions

Visualizations

Solver agnostic

Support for parallel computing

·

·

·

·

·

·

·

5/38


5 of 38

New in PortfolioAnalyticsPushed to CRAN

Regime Switching Framework

Multilayer Optimization

Rank Based Optimization

Factor Model Moment Estimates

Improved Random Portfolios Algorithm

More demos, vignettes, and documentation

·

·

·

·

·

·

·

6/38


6 of 38

Multilayer Optimization

7/38


7 of 38

Support Multiple SolversLinear and Quadratic Programming Solvers

Global (stochastic or continuous solvers)

R Optimization Infrastructure (ROI)·

GLPK (Rglpk)

Symphony (Rsymphony)

Quadprog (quadprog)

-

-

-

Random Portfolios

Di�erential Evolution (DEoptim)

Particle Swarm Optimization (pso)

Generalized Simulated Annealing (GenSA)

·

·

·

·

8/38


8 of 38

Random PortfoliosPortfolioAnalytics has three methods to generate random portfolios.

The sample method to generate random portfolios is based on an idea by Pat Burns.1.

The simplex method to generate random portfolios is based on a paper by W. T. Shaw.2.

The grid method to generate random portfolios is based on the gridSearch function in theNMOF package.

3.

9/38


9 of 38

Comparison of Random Portfolio Methods(Interactive!)

10/38

0.0100 0.0150 0.0200 0.0250 0.0300

Risk (std. dev.)

0.0082 0.0345

0.0055

0.0060

0.0065

0.0070

0.0075

Ret

urn

0.0051

0.0078samplesimplexgridMagnify


10 of 38

Random Portfolios: Simplex Method

11/38


11 of 38

Workflow: Specify Portfolio

Initializes the portfolio object that holds portfolio level data, constraints, and objectives

args(portfolio.spec)

## function (assets = NULL, category_labels = NULL, weight_seq = NULL,

## message = FALSE)

## NULL

12/38


12 of 38

Workflow: Add Constraints

Supported Constraint Types

args(add.constraint)

## function (portfolio, type, enabled = TRUE, message = FALSE, ...,

## indexnum = NULL)

## NULL

Sum of Weights

Box

Group

Factor Exposure

Position Limit

and many more

·

·

·

·

·

·

13/38


13 of 38

Workflow: Add Objectives

Supported Objective types

args(add.objective)

## function (portfolio, constraints = NULL, type, name, arguments = NULL,

## enabled = TRUE, ..., indexnum = NULL)

## NULL

Return

Risk

Risk Budget

Weight Concentration

·

·

·

·

14/38


14 of 38

Workflow: Run Optimizationargs(optimize.portfolio)

## function (R, portfolio = NULL, constraints = NULL, objectives = NULL,

## optimize_method = c("DEoptim", "random", "ROI", "pso", "GenSA"),

## search_size = 20000, trace = FALSE, ..., rp = NULL, momentFUN = "set.portfolio.moments",

## message = FALSE)

## NULL

args(optimize.portfolio.rebalancing)

## function (R, portfolio = NULL, constraints = NULL, objectives = NULL,

## optimize_method = c("DEoptim", "random", "ROI"), search_size = 20000,

## trace = FALSE, ..., rp = NULL, rebalance_on = NULL, training_period = NULL,

## rolling_window = NULL)

## NULL

15/38


15 of 38

Workflow: Analyze ResultsVISUALIZATION DATA EXTRACTION

plot extractObjectiveMeasures

chart.Concentration extractStats

chart.EfficientFrontier extractWeights

chart.RiskReward print

chart.RiskBudget summary

chart.Weights

16/38


16 of 38

Portfolio Optimization

17/38


17 of 38

Estimating MomentsLedoit and Wolf (2003):

"The central message of this paper is that nobody should be using the sample covariancematrix for the purpose of portfolio optimization."

Sample

Shrinkage Estimators

Factor Model

Expressing Views

·

·

·

·

18/38


18 of 38

Meucci Fully Flexible ViewsReference Model

Views

Express View on Ranking

Posterior

·

X ∼ fX

·

V ≡ g(X) ∼ fV

·

m{ } ≥ m{ } ≥. . . ≥ m{ }V1 V2 VK

·

≡ entropy(f , )f̃ x argmin f∈V

fx

19/38


19 of 38

Almgren-Chriss Portfolios from SortsDe�ne as the investment universe of assets

De�ning Sorts

Centroid vector, , is de�ned as the center of mass of the set .

· , , . . . ,S1 S2 Sn n

·

Single complete sort

Sector based sort

Deciles and other divisions

Single complete sort with longs and shorts

others

-

≥ ≥. . . ≥r1 r2 rn

-

-

-

-

· c Q

where is the space of consistent expected returns- Q

20/38


20 of 38

Example 1 and 2Consider an allocation to hedge funds using the EDHEC-Risk Alternative Index as a proxy.

Simple example of expressing a view on the order of the expected returns of assets usingboth Meucci's Fully Flexible Views and the Almgren-Chriss framework

Establish a quantitative rule for our view on the order of the expected returns of assets andtest on historical data.

·

·

21/38


21 of 38

Example 1: Data and Portfolio# Load package and data

library(PortfolioAnalytics)

source("data_prep.R")

R <- edhec[,1:4]

funds <- colnames(R)

# Construct portfolio

meanSD.portf <- portfolio.spec(assets=funds)

meanSD.portf <- add.constraint(portfolio=meanSD.portf, type="weight_sum",

min_sum=0.99, max_sum=1.01)

meanSD.portf <- add.constraint(portfolio=meanSD.portf, type="box",

min=0.05, max=0.5)

meanSD.portf <- add.objective(portfolio=meanSD.portf, type="risk", name="StdDev")

meanSD.portf <- add.objective(portfolio=meanSD.portf, type="return", name="mean")

# Generate random portfolios for use in the optimization.

rp <- random_portfolios(meanSD.portf, 5000)

22/38


22 of 38

Example 1: Express Views

Meucci: Fully Flexible Views Framework

Almgren and Chriss: Portfolios from Sorts

# Here we express views on the relative rank of the asset returns

# E{ R[,2] < R[,3] < R[,1] < R[,4] }

asset.rank <- c(2, 3, 1, 4)

p <- rep(1 / nrow(R), nrow(R))

m.moments <- meucci.ranking(R, p, asset.rank)

ac.moments <- list()

ac.moments$mu <- ac.ranking(R, asset.rank)

# Sample estimate for second moment

ac.moments$sigma <- cov(R)

23/38


23 of 38

Example 1: Optimization# Use moments output from meucci.ranking

opt.meucci <- optimize.portfolio(R, portfolio=meanSD.portf,

optimize_method="random",

rp=rp,

trace=TRUE,

momentargs=m.moments)

# Use first moment from ac.ranking. Note second moment is sample covariance

opt.ac <- optimize.portfolio(R, portfolio=meanSD.portf,

optimize_method="random",

rp=rp,

trace=TRUE,

momentargs=ac.moments)

24/38


24 of 38

Example 1: Optimization Results Optimal Weights

25/38


25 of 38

Example 2: Custom Moment Functionmoment.ranking <- function(R, n=1, method=c("meucci", "ac")){

method <- match.arg(method)

tmpR <- apply(tail(R, n), 2, function(x) prod(1 + x) - 1)

# Assume that the assets with the highest return will continue to outperform

asset.rank <- order(tmpR)

switch(method,

meucci = {

p <- rep(1 / nrow(R), nrow(R))

moments <- meucci.ranking(R, p, asset.rank)

},

ac = {

moments <- list()

moments$mu <- ac.ranking(R, asset.rank)

moments$sigma <- cov(R)

})

moments

}

26/38


26 of 38

Example 2: Optimization with Periodic Rebalancingopt.bt.meucci <- optimize.portfolio.rebalancing(R, portfolio=meanSD.portf,

optimize_method="random", rp=rp,

rebalance_on="quarters",

training_period=72,

momentFUN="moment.ranking",

n=3, method="meucci")

opt.bt.ac <- optimize.portfolio.rebalancing(R, portfolio=meanSD.portf,



training_period=72,

momentFUN="moment.ranking",

n=3, method="ac")

27/38


27 of 38

Example 2: Optimization Results

28/38


28 of 38

Example 3: Minimum Expected Shortfall with RiskBudget LimitConsider an allocation to equity sectors using 9 sector ETFs.

Raw returns data and sample moment estimates

Cleaned returns data and factor model moment estimates

·

·

29/38


29 of 38

Example 3: Data and Portfolio# Data

R.raw <- ret.sector

R <- Return.clean(R.raw, "boudt")

funds <- colnames(R)

# Construct initial portfolio with basic constraints.

ES.portf <- portfolio.spec(assets=funds)

ES.portf <- add.constraint(portfolio=ES.portf, type="weight_sum",

min_sum=0.99, max_sum=1.01)

ES.portf <- add.constraint(portfolio=ES.portf, type="long_only")

ES.portf <- add.objective(portfolio=ES.portf, type="risk", name="ES",

arguments=list(p=0.95))

ES.portf <- add.objective(portfolio=ES.portf, type="risk_budget",

name="ES", max_prisk=0.25,

arguments=list(p=0.95))

# Generate random portfolios

rp <- random_portfolios(ES.portf, 5000)

30/38


30 of 38

Example 3: Fit Statistical Factor Model# This is not necessary for the optimization, but demonstrates how to extract

# the moments for use in custom objective function

fit <- statistical.factor.model(R, k=3)

# Extract the moments

sigma <- extractCovariance(fit)

m3 <- extractCoskewness(fit)

m4 <- extractCokurtosis(fit)

31/38


31 of 38

Example 3: Custom Moment Functionfm.moments <- function(R, k=1){

fit <- statistical.factor.model(R=R, k=k)

momentargs <- list()

momentargs$mu <- matrix(rep(0, ncol(R)),ncol=1)

momentargs$sigma <- extractCovariance(fit)

momentargs$m3 <- extractCoskewness(fit)

momentargs$m4 <- extractCokurtosis(fit)

return(momentargs)

}

32/38


32 of 38

Example 3: Optimization with Periodic Rebalancing# Higher moment estimates from statistical factor model

minES.boudt <- optimize.portfolio.rebalancing(R=R, portfolio=ES.portf,

momentFUN=fm.moments, k=3,



training_period=1250,

trace=TRUE)

# Sample estimates of the moments

minES.sample <- optimize.portfolio.rebalancing(R=R.raw, portfolio=ES.portf,



training_period=1250,

trace=TRUE)

33/38


33 of 38

Example 3: Optimization Results

34/38


34 of 38

Conclusion

AcknowledgementsMany thanks to...

Introduced the goals and summary of PortfolioAnalytics

Demonstrated the �exibility through examples

Plans for continued development

·

·

·

Interface to

Additional solvers

"Gallery" of examples

- parma-

-

Google: funding Google Summer of Code (GSoC) for 2013 and 2014

UW CF&RM Program: continued work on PortfolioAnalytics

GSoC Mentors: Brian Peterson, Peter Carl, Doug Martin, and Guy Yollin

R/Finance Committee

·

·

·

·

35/38


35 of 38

PortfolioAnalytics Links

Source code for the slides

and view it here

PortfolioAnalytics on CRAN

PortfolioAnalytics on R-Forge

·

·

https://github.com/rossb34/PortfolioAnalyticsPresentation2015·

http://rossb34.github.io/PortfolioAnalyticsPresentation2015/·

36/38


36 of 38

Any Questions?

37/38


37 of 38

References and Useful LinksROI

DEoptim

pso

GenSA

PerformanceAnalytics

Patrick Burns Random Portfolios

W.T. Shaw Random Portfolios

Improved Forecasts of Higher-Order Co-moments and Implications for Portfolio Selection

Higher Order Comoments of Multifactor Models and Asset Allocation

Shiny App

·

·

·

·

·

·

·

·

·

·

38/38


38 of 38

Documents

Advanced Moment Estimation & Optimization with …past.rinfinance.com/agenda/2015/workshop/RossBennett.pdf · 2015-06-08 · PortfolioAnalytics is an R package designed to provide